Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

On the Similarity of Deep Learning Representations Across Didactic and Adversarial Examples

Pk Douglas, Farzad Vasheghani Farahani

TL;DR

It is shown that representational similarity and performance vary according to the frequency of adversarial examples in the input space, which indicates the stability of deep learning representations for neuroimaging classification across didactic and adversarial conditions characteristic of MRI acquisition variability.

Abstract

The increasing use of deep neural networks (DNNs) has motivated a parallel endeavor: the design of adversaries that profit from successful misclassifications. However, not all adversarial examples are crafted for malicious purposes. For example, real world systems often contain physical, temporal, and sampling variability across instrumentation. Adversarial examples in the wild may inadvertently prove deleterious for accurate predictive modeling. Conversely, naturally occurring covariance of image features may serve didactic purposes. Here, we studied the stability of deep learning representations for neuroimaging classification across didactic and adversarial conditions characteristic of MRI acquisition variability. We show that representational similarity and performance vary according to the frequency of adversarial examples in the input space.

On the Similarity of Deep Learning Representations Across Didactic and Adversarial Examples

TL;DR

It is shown that representational similarity and performance vary according to the frequency of adversarial examples in the input space, which indicates the stability of deep learning representations for neuroimaging classification across didactic and adversarial conditions characteristic of MRI acquisition variability.

Abstract

The increasing use of deep neural networks (DNNs) has motivated a parallel endeavor: the design of adversaries that profit from successful misclassifications. However, not all adversarial examples are crafted for malicious purposes. For example, real world systems often contain physical, temporal, and sampling variability across instrumentation. Adversarial examples in the wild may inadvertently prove deleterious for accurate predictive modeling. Conversely, naturally occurring covariance of image features may serve didactic purposes. Here, we studied the stability of deep learning representations for neuroimaging classification across didactic and adversarial conditions characteristic of MRI acquisition variability. We show that representational similarity and performance vary according to the frequency of adversarial examples in the input space.

Paper Structure

This paper contains 6 sections, 1 equation, 2 figures.

Table of Contents

  1. Introduction
  2. Related work
  3. Deep Learning with Adversarial and Didactic Examples
  4. Relevance Structural Similarity Analysis
  5. Conclusions
  6. Acknowledgments

Figures (2)

  • Figure 1: (Top) Adversarial exemplars $\tilde{X}$ were created with the addition of noise found commonly in the MR setting. (Middle) Didactic examples were created by placing an image (a panda) in a location that varied according to the class label. The image was added to the top right non-brain region for ADHD subjects and top left for TD. (Lower from left to right) Accuracy across different CNN models using original data; accuracy when percentage of noise exemplars varies, shown for Rician noise $\lambda$=0.15 ; Loss over training epochs when all exemplars were didactic.
  • Figure 2: Relevance Structural Similarity Analysis (RSSA) results. (Top Left) RSS across Rician noise levels. (Top Middle, Right) RSSA matrix across each noise type compared to original data. (Bottom) RSSA maps for LIME and LRP for Rician noise and didactic examples.